A discontinued Android curio with no apps, no recovery, and a sealed bootloader — turned into a dual-core postmarketOS media player with Spotify Connect, Bluetooth A2DP, a beat-reactive 32-LED ring, an on-demand Wayland desktop you can drive with a BT mouse + keyboard, a 1.2 GHz CPU, NFC tap-to-send, and a phone/desktop companion remote that doubles as the screenless orb's Bluetooth settings panel.
The Nexus Q (codename steelhead) was Google's mysterious 2012 media sphere:
a TI OMAP4460, a 25 W amplifier, a ring of 32 RGB LEDs, and an Android build
that did almost nothing. Google cancelled it before it ever really shipped.
Nexus Q Reloaded throws away the Android stack and boots a mainline Linux 6.12 LTS kernel under postmarketOS — reverse-engineering the factory kernel where mainline fell short, and bringing the orb back as something genuinely useful.
It plays music. It glows in time. It runs
python3,ssh, and a desktop. On a phone from before the original was even released.
| Subsystem | Status | Notes |
|---|---|---|
| 🐧 Boot — mainline 6.12 + postmarketOS (systemd) | ✅ | daily-usable from a clean flash · genuinely clean boot log — 0 failed units, dmesg err/warn EMPTY, and journalctl -b -p warning down to only 4 documented-external lines (all ~15 v1.6.9 residual err/warn lines root-caused + fixed · was 3 externals v1.6.10–v1.8.1; a 4th — a one-shot NM vendored-libsystemd assert at the RTC→NTP clock jump — was dispositioned 2026-07-13) · v1.6.10 |
| ⚡ Dual-core SMP | ✅ | both Cortex-A9 cores online (nproc=2) · since v1.2.0 |
| 🚄 CPU freq scaling 350 → 1200 MHz | ✅ | DVFS · v1.4.0 · governor conservative since v1.8.2 — a measured 2026-07-13 idle study showed ondemand kept 74 % of idle at ≥700 MHz on microburst wakeups (~1000/s); conservative won the A/B/C test and idle now settles at 350 MHz (56.7 % residency, 4.25 trans/s). History: conservative v1.5.0–v1.6.5 → ondemand v1.6.6–v1.8.1 → conservative v1.8.2 (this time measurement-backed) |
| 🔊 TAS5713 25 W speaker | ✅ | audible since v1.6.13 (kernel r36). The software pipeline (driver/PCM/softvol, correct pitch — 2× clock bug) landed v1.6.1, but the physical amp was silent through every earlier release: mcbsp2_pins muxed the wrong balls (abe_dmic_*), so the McBSP2 I2S clock/data/frame never reached the amp (aplay rc=0, nothing driven). Root-caused + fixed in DTS 2026-07-07 (stock pads 0x0f6/0x0fa/0x0fc MUX_MODE0) → user-confirmed audible. Now one selectable PulseAudio output (v1.6.15, shipped in v1.7.0). The residual playback crackle is CLOSED 2026-07-12 — it was TWO independent faults, both fixed (hardware-verified, user-confirmed perfectly clean playback): (a) load-correlated bus/DMA contention → kernel r41 patch 0041 (sDMA CCR_READ_PRIORITY on the cyclic audio channel + GCR HI_THREAD_RESERVED=1; verified GCR=0x00011010, ch20 CCR bit6=1); (b) a metronomic ~1/s click from two free-running crystals — mainline reparented the DPLL_ABE reference to sys_32k while the TAS5713 MCLK sat on the 38.4 MHz crystal (~21 ppm ≈ 1 sample slip/s @ 48 kHz) → kernel r42 patch 0042 relocks DPLL_ABE from sys_clkin at exactly 98.304 MHz, the stock topology the bootloader sets and our port was undoing. See docs/2026-07-12-audio-crackle-closed-sdma-priority-and-dpll-abe.md (+ the 07-08/07-09 diagnosis notes) |
| 🎵 Spotify Connect (librespot) | ✅ | advertises "Nexus Q", streams over 5 GHz · v1.6.1 · now a PulseAudio input (systemd user unit → --device pulse), one movable PA sink-input · v1.6.15 |
| 🍏 AirPlay (shairport-sync) | ✅ | a PulseAudio input like librespot (user unit, alsa→pulse), advertises the onboarding name via avahi (_raop._tcp), ports pinned (RTSP 5000 + UDP 6001-6010, 61_airplay.nft) · user-tested playing 2026-07-17 · device r50 · in-source, first flashed image pending (v1.11.0-rc1) |
| 🎼 Roon Bridge (Roon Ready endpoint) | ✅ | Roon's glibc/Mono binaries can't run on musl (gcompat segfaults them), so the Bridge runs in a bwrap sandbox over a baked Debian-armhf glibc base (/opt/glibc-rt) — only the base is baked, RoonBridge is lazy-fetched on first use and self-updates. Audio via a dedicated 2nd snd-aloop card → PA roon_in → default sink, so it follows the output selector like every input. Validated end-to-end against a real ROCK Core 2026-07-17 (discovery, enable, playback). Default-OFF user unit — enabled on demand (systemctl --user enable --now roon), the resource policy (companion-app per-service toggles are the next feature). See docs/2026-07-17-roon-bring-up.md · device r55 · in-source, first flashed image pending. Tidal deferred (grey-area binary) |
| 🔊 Audio output selection (speaker / optical / HDMI) | ✅ | v1.6.15 (shipped in v1.7.0): PulseAudio is the hub, the active output = the PA default sink, picked from the companion app (listOutputs/setOutput → pactl set-default-sink + move all sink-inputs + class-D amp safety toggle). Input-agnostic + future-proof (BT-A2DP / Tidal / casting can join as further PA inputs) |
| 🔴 LED music visualizer | ✅ | the ring dances to the beat · v1.6.2 · 5 selectable visualisations + breathing color themes · idle-keepalive (no more dark-after-idle AVR starvation) · v1.6.5 · volume-independent — re-tapped to the active output's PA monitor + an AGC (auto-gain) so it reacts to the music at any listening volume, no low-volume flicker · v1.6.15 · tap now gated on playback so the amp sink suspends when idle (idle CPU ~7 % → ~1 %) · v1.7.1 |
| 📱 Companion app + LAN control bridge | ✅ | Flutter remote → nexusq-control (TCP 45015, mDNS): volume · breathing LED theme + brightness · visualisation picker · now-playing · v1.6.3 · reachable over WiFi · v1.6.5 · output selector (Holo-dark segmented control) + volume/mute now act on the active PA sink · v1.6.15 · NFC tap-to-send receiver (HCE — the Q taps a message onto the phone, shown as a SnackBar) + auto-reconnect on resume/drop (no more app-kill after backgrounding) · v1.7.0 · two-way volume sync — the app slider now tracks the physical dome dial and the LXQt applet (bridge pactl subscribe → volumeChanged) · v1.7.3 (verified live, not yet in a flashed image) · Devices screen (app 1.2.0+7, device v1.10.0): Pair a phone / Add a mouse or keyboard / paired list with Forget / HDMI desktop toggle, from the home app bar. |
| 🖥 HDMI desktop (LXQt · Wayland) · on demand | ✅ | toggle-able from the app since v1.10.0 (setDesktop/getDesktop → tinydm.service) — the desktop idles the GPU/display path and heats the sphere, so it is on-request. The user linger is load-bearing (baked in device r48): PA + librespot are user units under user@10000.service, the desktop is tinydm → labwc in session-c1.scope — without linger the user manager exists only because of the graphical session, so stopping the desktop would kill the music. Verified: with linger, systemctl stop tinydm leaves pulseaudio + librespot active, both sinks present. Pair a keyboard + mouse (row above), switch the desktop on → the appliance is a computer. · labwc + Pixman renderer · desktop audio sink fixed v1.6.12 (the red-cross no-sink tray icon: PA now starts via a native systemd USER unit — Alpine ships none and the XDG autostart never fires under systemd+Wayland — and the sole sink is the TAS5713 speaker) |
| 📶 WiFi (BCM4330, 5 GHz) | ✅ | NetworkManager, factory MAC f8:8f:ca:20:48:e1 pinned in the DTS since v1.10.1 (kernel patch 0043 — local-mac-address, mirrors the BT node; ethtool -P wlan0 reports it as PERMANENT, so every profile incl. the onboarding one gets it — the old NM cloned-mac-address pin only reached the baked profile, and ≤v1.10.0 fell back to the chip OTP MAC 14:7d:c5:3a:35:b5). The router can still reassign the DHCP lease (seen 2026-07-12): find the device by hostname steelhead/MAC, don't hardcode the IP. Characterized 2026-07-07: 5 GHz is healthy — NOT flaky (−48 dBm, 0 discarded/retry pkts, 2.6 ms jitter, 0 % loss); bulk ~34 Mbit/s is a hardware ceiling of the 2010-era 1×1 802.11n BCM4330 (not a bug — same cipher does ~80 over ethernet, so WiFi is the limit; ~100× the appliance's need). Use ethernet for bulk |
| 🔵 Bluetooth + A2DP audio (BCM4330) | ✅ | A2DP sink reliable since v1.8.0 — pair a phone and stream to the Q (phone → BT → PulseAudio bluez_source s24le/48 kHz → TAS5713). Root cause of every past "won't stay connected / phantom Connected / corrupt-burst audio" was a missing BT HCI UART max-speed: the BCM4330 HCI runs over UART2 and hci_bcm left oper_speed=0, never syncing the host UART to the firmware baud → hci0: Frame reassembly failed (-84) (EILSEQ) + tx timeouts. Kernel patch 0040 sets max-speed = <3000000> (stock ran 3 Mbaud); verified live — reassembly failures 0 (was 26+), controller addr correct. (NOT coexistence, NOT HFP/SCO — both earlier wrong guesses.) Pairing is NoInputNoOutput Just-Works via nexusq-btagent (the Q's single permanent agent — nothing attached to it can answer a prompt; bonds are marked Trusted, and the ring spins blue ⇔ the Q is pairable). blueman-applet must stay out of the session: its DisplayYesNo agent forces SSP into Numeric Comparison → an unanswerable HDMI dialog → every bond times out (root-caused 2026-07-15; suppressed since device r47 — the package stays). Per-device BD_ADDR F8:8F:CA:20:49:E5 since v1.6.10 (DTS local-bd-address + btbcm patch 0036) |
| 🖱 BT pairing from the app — both directions | ✅ | v1.10.0 — the Q has no screen and no input device, so the app IS the Q's Bluetooth settings panel; there is no other way to pair anything to it. Inbound: a phone pairs for music (A2DP). Outbound: the Q scans for and pairs a mouse / keyboard — a different flow, not a variant (a mouse never connects TO us; we must discover it and call Pair() on it). Hardware-verified 2026-07-15: pairBtDevice → {"paired":true,"bonded":true,"connected":true}, 3 key sections on disk, kernel created MX Master 4 Mouse on /dev/input/… via uhid; a real BLE keyboard (MX Keys) completes Just Works against our NoInputNoOutput agent with no typed passkey. bonded, not paired, is the honest answer to "will this survive a reboot?" — paired alone LIES. Root cause this release is built on: v1.9.0's Pairable == Discoverable invariant was keyed on the WRONG property and silently broke OUTBOUND bonding (Pairable: no → pair "succeeds", no keys stored, gone on restart; Pairable: yes → [PeripheralLongTermKey] + [IdentityResolvingKey] on disk, survives). Chain measured from bluetoothd -d: the LTK arrives, but bluez only persists a key the kernel marked store_hint, which needs the SMP bonding bit, which our side only sets under HCI_BONDABLE = Adapter1.Pairable. Now: ring ⇔ Pairable, off at rest, and an outbound pair opens a window like everything else. See docs/2026-07-15-step2-bt-pairing-implemented.md + PROTOCOL §9 |
| 📲 App-driven onboarding (NFC tap → BT → WiFi) | ✅ | works end-to-end from a fresh flash · v1.9.0 — hardware-accepted 2026-07-15. Tap the phone on the dome → bonded, encrypted BT RFCOMM (nexusq-setupd, RequireAuthentication=True — the WiFi PSK never crosses the air in cleartext) → WiFi join → name/room/theme → outro, with the original stock imagery. Final acceptance on a fresh v1.9.0-rc5 flash: tap delivered → bond first try (0 failed attempts) → RFCOMM → WiFi joined → finishSetup → pairing window auto-closes (PSK: 0 log lines); wrong password → ring turns red. The pairing window fails CLOSED (a transient NetworkManager wobble can no longer drop a provisioned device into a discoverable+pairable setup mode). startSetupMode re-provisioning tested + passing. PUBLIC_RELEASE=1 images do onboard). See docs/2026-07-15-bt-onboarding-root-caused-blueman-agent-and-bond-first.md |
| 🔐 SSH (USB-gadget + WiFi) | ✅ | RNDIS net 172.16.42.1 + ACM console. On v1.6.5 only user@ works; key-based root@ is baked in + verified 2026-07-03 (ships in v1.6.6) |
| 🐍 python3 on-device | ✅ | flash-verified · v1.6.0 |
| 🌡 TMP101 temperature sensor | ✅ | |
| 📡 NFC tap-to-send (PN544) | ✅ | tap-to-send shipped v1.7.0 (2026-07-08, verified on device): tap a phone on the dome → the Q pushes a short text over NFC, shown in the companion app. Reverse-HCE — the PN544 can't host-card-emulate (no SE) and Android Beam is gone, so the phone runs the HCE service and the Q is the ISO-DEP reader (nexusq-nfc-send daemon, AID F0010203040506). Key enabler: kernel patch 0037 RATS-activates any ISO-DEP target (was DESFire-only), so a modern HCE phone (SAK 0x20) is finally reachable. The chip itself was fixed 2026-07-03 (v1.6.6) — the DTS had muxed the wrong pads (dpm_emu debug pads instead of usbb2_ulpitll_dat1/2/3), found via a stock RAM-boot probe. See docs/2026-07-08-nfc-tap-to-send-reverse-hce.md |
| 🔈 HDMI audio | 🟠 | needs a sink with audio EDID (the card is a dummy-DAI — PA ignores it via a PULSE_IGNORE udev rule, so no more boot-log noise · v1.6.9); joins the output selector as hdmi once that rule is lifted against a real audio sink (TV/AVR) — UNTESTED |
| 🌐 Ethernet (LAN9500A) | ✅ | works from a cold boot — task #17 fully closed (gold-validated: clean flash + true cold power-cycle → eth0 100Mbps/Full, 0 failed units). The "enumeration intermittency" was a pinmux miss: gpio_1 NENABLE (the LAN9500A power-enable) sat on an unmuxed pad (kpd_col2 @ padconf 0x186) so it never powered the chip — the healthy USB3320 PHY masked it, and the earlier "3/3 vs 0/3 boots" was stock priming, not a race. Fixed in kernel #33 (DTS pad mux); the 2500ms "settle" it superseded was a false positive · v1.6.8. NM layer resolved 2026-07-04 (baked eth-lan DHCP + eth-direct static ssh root@10.42.0.2). Now the DEFAULT deploy/control path (measured 2026-07-07: ~80 Mbit/s, 0.62 ms — faster + more stable than WiFi/USB-gadget, fixed IP; the direct-cable static profile auto-comes-up since device r29 · v1.6.12). Chip has no MAC EEPROM → random MAC/lease per boot on a LAN |
| 💿 TOSLINK / SPDIF | ✅ | brought up in v1.6.13 — no C driver (mainline davinci-mcasp DIT/IEC958): defconfig SND_SOC_DAVINCI_MCASP=m+SND_SOC_SPDIF=m, DTS &mcasp0 + mcasp_spdif_pins (0x0f8 MUX_MODE2, AXR0) + sound_spdif card. Probe -EINVAL fixed via format="i2s"+mcasp master. A selectable PA output ("Optický výstup") since v1.6.15; PA pinned to 48 kHz (50-nexusq-48k.conf) so the DIT locks (44.1 kHz → "off by 88435 PPM"). Both PA sinks report 48000 Hz on fresh boot |
| 🎧 TWL6040 headset codec | ⚪ | not populated/unused on steelhead — the stock kernel never drove it (verified 2026-07-03); no headset path by design (was wrongly called "dead hardware") |
Full per-milestone detail in CHANGELOG.md · hardware map & roadmap in PLAN.md.
How a tap on your phone becomes sound and light — the heart of the v1.6.x work:
flowchart LR
P([📱 Phone<br/>Spotify app]) -->|mDNS · Spotify Connect| L[librespot<br/>“Nexus Q”]
P -.->|🔵 Bluetooth A2DP · v1.8.0| B[bluez_source<br/>s24le · 48 kHz]
L -->|--device pulse| PA{{PulseAudio<br/>hub · 48 kHz}}
B -.->|loopback| PA
PA -->|default sink| S([🔊 TAS5713<br/>25 W speaker])
PA -.->|selectable| SP([💿 optical SPDIF])
PA -.->|selectable| HD([🔈 HDMI])
PA -->|sink.monitor · arecord| N[nexusqd<br/>FFT · beat · AGC]
N -->|I²C → AVR| R(((🔴 32-LED ring)))
style S fill:#1f6feb,stroke:#1f6feb,color:#fff
style R fill:#b62324,stroke:#b62324,color:#fff
style L fill:#1db954,stroke:#1db954,color:#fff
Since v1.6.15 PulseAudio is the hub: librespot feeds it as one input
(--device pulse), and the active output — TAS5713 speaker, optical SPDIF, or
HDMI — is the PA default sink, chosen from the companion app. The LED daemon reads
the active sink's monitor, runs an FFT with an auto-gain stage, and animates the
ring — so the orb glows in time with whatever you're playing, at any volume. (Before
v1.6.15 the stream was teed via an ALSA type multi to the amp + a snd-aloop
loopback; the McBSP2 pinmux fix in v1.6.13 was what first made the physical amp
audible at all.)
Since v1.6.3 a phone/desktop companion app auto-discovers the Q over mDNS and
controls volume (since v1.6.15 the active PA output's sink; input-agnostic), the
audio output (speaker / optical / HDMI · v1.6.15), the LED color theme +
brightness, the music visualisation, mute (with a device-side mute-LED indicator),
and shows now-playing — talking to the on-device nexusq-control LAN bridge (TCP 45015,
line-delimited JSON — reachable over WiFi since v1.6.5). Since v1.6.5 a color theme
is a breathing override (the ring gently pulses in the theme's hue, always visible) while a
separate picker chooses one of the 5 music-reactive visualisations shown while audio
plays. The Flutter app is installed on the phone, not in the device image.
Grab the latest release, then:
# 1. Enter fastboot: unplug power, cover the top mute-LED sensor with your palm,
# plug power back in. The ring turns solid red.
# 2. Decompress the rootfs and flash
zstd -d nexusq-rootfs-v*-sparse.img.zst
fastboot flash boot nexusq-boot-v*.img
fastboot -S 100M flash userdata nexusq-rootfs-v*-sparse.img # -S chunking is REQUIRED
# 3. Power-cycle without covering the sensor. Tux → kernel → desktop.Then open Spotify on the same WiFi and cast to "Nexus Q" 🎶. Full walkthrough in INSTALL.md.
| Component | Chip | Driver | Bus |
|---|---|---|---|
| SoC | TI OMAP4460 (Cortex-A9 ×2) | omap4 |
— |
| Audio amp | TI TAS5713 25 W Class-D | snd-soc-tas571x |
McBSP2 / I²C4 |
| Audio codec | — (TWL6040 pad unpopulated/unused; stock never drove it) | none — removed from DTS/defconfig | — |
| WiFi | Broadcom BCM4330 | brcmfmac |
SDIO / MMC5 |
| Bluetooth | Broadcom BCM4330 | hci_bcm |
UART2 |
| NFC | NXP PN544 | pn544_i2c |
I²C3 |
| Ethernet | SMSC LAN9500A | smsc95xx |
USB EHCI |
| HDMI | OMAP4 DSS + TPD12S015A | omapdrm |
DSS |
| LED ring | AVR MCU (32 RGB) | leds-steelhead-avr |
I²C2 |
| PMIC | TI TWL6030 | twl-core |
I²C1 |
One command, fully dockerized (pmbootstrap under the hood):
./docker-build.sh # → output/boot.img + output/google-steelhead.imgIt builds the kernel (mainline 6.12.12 + 43 patches in kernel/patches/), the
local python3 override, the device daemons (nexusqd · nexusq-control ·
nexusq-btagent · nexusq-setupd), and a full systemd rootfs, then repacks a
ramdisk-less boot image and verifies the result by mounting it. Build notes and
the hard-won gotchas live in HANDOFF.md. (nexusq-btagent must build before nexusq-setupd, which depends
on it — the reverse order fails every clean build on checksums.)
kernel/ dts · defconfig · 43 mainline patches (the DTS ships VIA the patches — edit a patch, not just kernel/dts/)
pmos/ device-google-steelhead · linux-google-steelhead · firmware · nexusqd · nexusq-control · nexusq-btagent · nexusq-setupd · python3
userspace/ nexusqd (LED-ring daemon) · nexusq-control (LAN bridge) · nexusq-btagent (BT pairing agent) · nexusq-setupd (BT WiFi provisioning)
companion/ Flutter companion app + PROTOCOL.md (built on the phone, not in the image)
reverse-eng/ ground truth extracted from the factory kernel
scripts/ diagnostics (nq-healthd, nq-collect, …)
docs/ dated engineering record
raw2simg.py byte-exact all-RAW Android-sparse converter
0.1.0 ── first full boot, HDMI, WiFi, LED ring 2026-06-10
1.1.0 ── ethernet alive 2026-06-22
1.2.0 ── ✦ dual-core SMP 2026-06-23
1.3.0 ── ethernet hardened 2026-06-24
1.4.0 ── ✦ cpufreq DVFS → 1.2 GHz 2026-06-26
1.5.0 ── first full host-built rootfs 2026-06-27
1.6.0 ── ✦ python3 on-device (the flash-bug saga) 2026-06-28
1.6.1 ── ✦ TAS5713 audio fixed + Spotify Connect baked in 2026-06-29
1.6.2 ── ✦ LED music visualizer reacts to playback 2026-06-30
1.6.3 ── ✦ companion app + LAN control bridge 2026-06-30
1.6.5 ── ✦ breathing themes + 5 visualisations · LED keepalive · companion/WiFi 2026-07-01
1.6.6 ── ✦ NFC fixed (pinmux) · boot-error cleanup · factory MAC on air 2026-07-04
1.6.7 ── ✦ baked ethernet NM profiles · led_static healthd guard 2026-07-05
1.6.8 ── ✦ ethernet works from cold — unmuxed NENABLE pad (task #17 closed) 2026-07-06
1.6.9 ── ✦ boot log clean — gkr-pam + HDMI-audio noise silenced 2026-07-06
1.6.10 ─ ✦ boot log GENUINELY clean — dmesg err/warn EMPTY (all ~15 lines fixed) 2026-07-06
1.6.13 ─ ✦ TAS5713 speaker finally AUDIBLE (McBSP2 pinmux) + SPDIF bring-up 2026-07-07
1.6.15 ─ ✦ PA-centric audio: multi-input → PulseAudio → app-selectable output · LED AGC 2026-07-07
1.6.16 ─ ✦ physical volume dial → PulseAudio + tray icon follows output 2026-07-07
1.7.0 ── ✦ NFC tap-to-send (reverse-HCE, Q → phone) · companion auto-reconnect 2026-07-08
1.8.0 ── ✦ Bluetooth A2DP reliable (BT UART max-speed, patch 0040) · crackle isolated to output path 2026-07-10
1.8.1 ── ✦ playback crackle CLOSED — sDMA read-priority (r41) + DPLL_ABE sys_clkin relock (r42) hardware-verified 2026-07-12
1.8.2 ── ✦ idle power — conservative governor + healthd/pid-1 churn fixes (idle settles at 350 MHz) 2026-07-13
1.9.0 ── ✦ app-driven onboarding (NFC → bonded BT → WiFi) · pairing window fails CLOSED 2026-07-15
1.10.0 ─ ✦ BT pairing from the app, BOTH directions (phone in · mouse/keyboard out) · HDMI desktop on demand 2026-07-15
1.10.1 ─ ✦ bug-fix: factory WiFi MAC pinned in DT (patch 0043) · btagent fd leak · onboard · librespot boot race · app debug mode 2026-07-16 ← latest release
┊
(dev) ─ ✦ step 3 · streaming services: AirPlay (shairport-sync) · first-boot rootfs resize · Roon Bridge (glibc/bwrap, validated vs a real Core) 2026-07-17 ← in source (r55), v1.11.0-rc1 pending
(v1.7.4 was an unusable crackle-bake artifact — never shipped; v1.8.0 is its working successor.)
Where it started: Tux and a mainline 6.12 kernel reaching the Nexus Q's HDMI output
(an early 2026 milestone — the root filesystem came a few commits later).
GPL-2.0 — this repository carries Linux kernel patches, a device tree, and a defconfig, all derivative works of the Linux kernel (GPLv2).
Built with stubbornness for a sphere that deserved better. 🛸